Geometry of unimodular systems

A collection of vectors in a real vector space is called a unimodular system if any of its maximal linearly independent subsets generates the same free abelian group. This notion is closely connected with totally unimodular matrices: rows or columns of a totally unimodular matrix form a unimodular system and the matrix of coefficients of expansions of all vectors of a unimodular system with respect to its maximal linearly independent subset is totally unimodular. In this paper we show that a unimodular system defines the following geometric data: a Euclidean space, an integral lattice in it, and a reflexive lattice zonotope. The discriminant of the lattice is equal to the number of maximal linearly independent subsystems, and we call this number the complexity of the unimodular system. For a unimodular system $\Omega $ we also define the Gale dual unimodular system $\Omega ^{\bot}$ which has the same complexity. These notions may be illustrated by the well-known graphic and cographic unimodular systems of a graph. Both graphic and cographic unimodular systems have the same complexity which is equal to the complexity of the graph. For graphs without loops and bridges the graphic and the cographic unimodular systems are Gale dual to each other. We describe this geometric data for certain examples: for the graphic and the cographic unimodular systems of a generalized theta-graph, consisting of two vertices connected by $N$ edges, for the cographic system of the complete graph $K_N$, and for the famous Bixby-Seymour unimodular system, which is neither graphic nor cographic